The Berkshire of the agent era owns operations, not equities

Buffett’s genius was a structure, not a stock pick. Buy good businesses, leave the managers alone, route the cash they throw off into the next good business, and let compounding do the rest for fifty years. The holding company never ran the companies. It allocated capital between them and stayed out of the way. That was the right design, because running ten businesses by hand needed ten of everything: ten finance teams, ten ops leads, ten people who knew how the place actually worked.

That constraint is the thing that just broke.

The next great holding company won’t buy businesses and leave them alone. It will run a studio of small companies on one shared brain and one operating system, and the moat won’t be the balance sheet. It will be the OS.

The old holding company allocated capital because it couldn’t allocate anything else

Start with why the classic model looks the way it does. A holding company owns many businesses, and the one thing it can move cheaply between them is money. Cash from the furniture maker funds the insurer; the insurer’s float funds the railroad. Capital is liquid, fungible, and easy to point at the highest return. So the holding company became a capital allocator, and a brilliant one.

But look at everything it couldn’t move. It couldn’t take the furniture company’s best operator and have them also run finance for the insurer, that person can only be in one place. It couldn’t take the railroad’s hard-won knowledge of how to negotiate a supplier and instantly install it at the candy company. It couldn’t share a single accounting team across twelve subsidiaries, because the work was local, manual, and tied to people who lived inside one business.

So the answer was: own the businesses, share the capital, and keep everything else separate. Twelve companies meant twelve of every operational function. The holding company didn’t run operations across its portfolio because operations didn’t travel. People don’t fork. Knowledge didn’t copy. The only thing that crossed the boundary cleanly was money.

That wasn’t a philosophy. It was a limit. “Allocate capital, leave management alone” is the strategy you adopt when management is the one thing you can’t replicate.

What changes when operations becomes software

Now change the one variable. Suppose the operational layer of a company, the finance close, the inbox triage, the CRM that stays current, the SDR follow-ups, the weekly report nobody has time to write, isn’t a team of people anymore. Suppose it’s an operating system: agents that do the work, a memory that holds the context, a scheduler that runs the recurring jobs without being asked.

Software has a property people don’t have. It copies for free.

The naive read of “AI runs the company” is one super-app you point at a business and it handles everything. That’s the demo, and it falls over the moment you have two businesses, because each one has its own customers, its own data, its own rules, and its own definition of “done.” A single shared instance either leaks one company’s context into another’s or collapses into a lowest-common-denominator tool that fits none of them. The thing that kills a multi-business operator is exactly the thing the cloud spent a decade solving: how do you run one system for many tenants without their realities bleeding into each other?

The answer is the same one every serious platform reached. You don’t run one company on the OS. You run one instance per company on one shared OS, separate brains, separate data, separate agents, but the same engine underneath. The studio operates ten businesses the way a phone runs ten apps: isolated where it counts, shared where it pays.

And the moment operations runs on a shared OS, the boundary that used to block everything except money becomes the most permeable boundary in the company. The finance close that one business figured out is a pattern the OS can run for the next. The way the best instance handles a refund, a renewal, a cold lead, that’s no longer locked inside one team’s heads. It’s an agent behavior the studio can deploy across the whole portfolio the next morning.

The classic holding company allocated capital because capital was the only thing that crossed the boundary. The studio allocates operations, because now operations crosses too.

A studio of small companies, not a conglomerate of big ones

This flips the math on what size of business is worth owning.

The reason holding companies bought big businesses is that operational overhead is brutal at small scale. A company doing modest revenue still needs someone to do the books, someone to chase invoices, someone to answer support, someone to keep the CRM from rotting. Below a certain size, that overhead eats the whole margin, the business can’t carry the people it takes to run it. So the small business either stays a lifestyle operation or gets acquired and folded into something bigger to share the overhead. The minimum viable company was large because the operational floor was high.

Drop the floor and the math inverts.

Here’s the naive version of the inverted model, and why it’s not enough: “AI means one founder can run a business alone.” True, and it’s been said a hundred times. But one founder running one business is still bounded by that founder’s hours and attention. The interesting structure isn’t one person plus AI running one thing. It’s one studio, a small core of humans setting direction and taste, running a portfolio of small businesses, each operated by its own instance of the OS, none of which needs its own finance hire or ops lead because the OS is the finance hire and the ops lead, copied as many times as there are businesses.

Imagine a studio that owns, say, eight small companies, a niche e-commerce brand, a vertical software tool, a content property, a local service business, a few others. In the old world that’s eight management teams and a back office that would swallow any returns. In the studio world it’s eight instances of one OS, each running the recurring operational work, each holding its own memory of its own customers, all overseen by a handful of people who decide what each business should chase and what “great” means for the people it serves.

The unit of the portfolio shrinks because the cost of operating a unit collapsed. You stop buying the few big companies that can carry their own overhead, and start running many small ones that share it.

The moat is the operating system, not the portfolio

If you stop there, you’d think the advantage is just lower cost, cheaper to run, so you can run more. That’s real, but it’s not the moat. Cheaper operations is a discount anyone with the same tools can match. The moat is what the studio accumulates that a one-business operator can’t.

The naive view of competitive advantage in this world is “whoever has the best agents wins.” But agents are getting commoditized fast; the model under them improves for everyone at once. The durable edge isn’t the agents. It’s everything the studio learns from running many businesses through the same OS at the same time.

The next great holding company won’t buy businesses and leave them alone, its edge is that every business it runs makes the OS better at running the next one. Run eight businesses on one engine and you see eight versions of every operational problem: eight ways customers churn, eight ways a close goes wrong, eight ways a lead goes cold. The patterns that work in one become defaults in all. The failures caught in one become guards for the rest. A single business sees its own data and learns at the speed of its own mistakes. A studio sees the whole portfolio’s data and learns at the speed of all of them, and pours every lesson back into the shared engine.

That’s a compounding loop, and it’s the same shape that made the classic holding company great, just moved one layer down. Buffett’s cash compounded across the portfolio. The studio’s operational intelligence compounds across the portfolio. Every business funds the next one not only with cash but with learning. The OS that runs ten companies is a better OS than the one that ran nine, and the gap between it and a single-business tool widens with every company added.

A competitor can copy the idea of a studio. They can’t copy the years of accumulated operating behavior baked into an engine that has already run a hundred small businesses. That’s the moat the balance sheet used to be.

What the humans do when the OS runs the businesses

Here’s the part that isn’t about software.

If the OS does the operational work across every business in the portfolio, you’d expect the people to disappear. The opposite happens, the people get concentrated onto the only things that don’t copy. Someone still has to decide which businesses are worth owning. Someone has to define what “great” means for each one’s customers, because an OS can run a refund policy but it can’t decide whether the policy is generous or stingy, kind or cold. Someone has to hold the taste, the judgment about what each business should chase, what it should refuse, what it would be embarrassed to ship.

That’s not a smaller job. It’s the whole job, finally separated from the operational drag that used to bury it. The classic holding company concentrated a few brilliant people on capital allocation and let managers handle the rest. The studio concentrates a few people on taste and direction, across more businesses than any of them could ever have staffed, and lets the OS handle the rest.

The work that doesn’t fork is the work worth doing. Deciding what a company should be. Knowing the customer well enough to know what would delight them. Saying no to the business that isn’t worth owning and yes to the one that is. The OS frees the people in the studio to do only that, to be allocators of judgment instead of allocators of cash, across a portfolio that used to require an army and now requires a point of view.

---

That’s the bet we’re making at Apollo Space: not a smarter chatbot bolted onto one company, but an operating system clean enough to run many, each its own brain, its own data, its own agents, all on one engine that gets better every time it runs a new business. The Berkshire of the agent era won’t be a balance sheet that owns equities. It’ll be a piece of software that owns operations. If you’ve ever wanted to run ten companies with the team of one, the thing that was missing was never the ambition. It was the engine.